## 1. Effects of National Income on National Income Inequality (Baseline)

## Source details

**Canonical URL:** [1. Effects of National Income on National Income Inequality (Baseline)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14101.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14101.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14101.pdf.json)

---

### Summary of research question and approach
- Core questions:
  - Is income inequality reduced in the context of economic growth?
  - Does the relationship depend on the country’s level of economic development?
  - Are there non-linearities (e.g., a Kuznets inverted-U)?
  - What policy channel(s) mediate the inequality-reducing effect of income growth?
- Identification strategy:
  - Two instrumental variables isolate exogenous variation in national income over five-year horizons for a panel spanning 1960–2007.
  - IV1: oil price shocks (OPS) = Δln(OilPrice)_t * θ_c, where θ_c is the time-invariant average share of net oil exports in GDP for country c.
  - IV2: trade-weighted world income (TWWI) = ∑_{j≠i} w_ij GDP_jt, where w_ij is the share of trade between country i and country j in the GDP of country i (measured using trade shares between 1980 and 1989).
  - Main estimation: country fixed effects with year fixed effects using 5-year non-overlapping panel data; baseline sample period 1960–2007; final usable cross-country 5-year sample: 80 countries.
- Baseline regression:
  - Inequality_it = α_i + β_t + η ln(GDP p.c.)_it + u_it
  - Rationale: levels specification with country fixed effects assesses how within-country distribution changes as log national income changes.

### Principal causal findings: income → inequality
- Average effect:
  - A one percent increase in GDP per capita reduces the Gini coefficient by around 0.08 percentage points.
- Baseline 2SLS coefficients (Table 1, Panel A, 2SLS):
  - ln(GDP p.c.) on Gini = -8.48** (4.09)
  - Quintile effects (2SLS, Table 1 Panel A):
    - 1st Quintile: 1.43 (0.94)
    - 2nd Quintile: 1.70* (1.06)
    - 3rd Quintile: 2.01* (1.09)
    - 4th Quintile: 2.49*** (0.89)
    - 5th Quintile: -7.77** (3.52)
- Quantitative interpretation examples (preserving reported figures):
  - Estimated coefficient on log GDP per capita = -8.5 (example in text): a one percent (0.01 log point) increase in GDP per capita leads to a reduction in the Gini coefficient of over 0.08 percentage points.
  - Malaysia example: sample average Gini = 0.46; sample average PPP GDP per capita = $5174; as income per capita doubles, Gini expected to decline to 0.38.
  - Philippines example: sample average Gini = 0.43; sample average PPP GDP per capita = $2769; as income per capita doubles, Gini expected to decline to 0.35.
- Distributional pattern:
  - Income growth boosts the relative income share of the lower quintile and the middle-class (1st–4th quintiles).
  - Income growth results in a decline in the income share accruing to the top quintile (5th quintile).
- Poverty:
  - Panel A (country FE only, 2SLS) poverty results (Table 9):
    - PovGap $1.25: ln(GDP p.c.) = -3.93** (1.80)
    - PovGap $2.00: ln(GDP p.c.) = -6.57*** (2.43)
    - PovCount $1.25: ln(GDP p.c.) = -9.78** (3.90)
    - PovCount $2.00: ln(GDP p.c.) = -11.16*** (3.34)

### Robustness, dynamics, and alternative datasets
- LS vs 2SLS:
  - LS estimates (Panel B) yield an insignificant relationship between national income and income inequality when controlling for country and year fixed effects.
  - Panel C (level and squared GDP per capita): coefficients insignificant; p-value testing equality of linear and squared terms always exceeds 0.1 — no Kuznets relationship found.
- Instrument diagnostics:
  - First-stage Kleibergen-Paap F-statistic well in excess of 10 in all regressions reported; Hansen J p-values "always in excess of 0.1" in reported specifications (examples: Table 1 Panel A Hansen J p-value 0.47).
  - Tests for exclusion restriction: cross-country correlation between average oil net-export GDP shares and Gini = 0.11; regression t-value 1.01 (cannot reject zero).
  - Conditional tests: OPS effects on inequality conditional on GDP per capita insignificant (no evidence of direct OPS → inequality beyond GDP channel).
- Dynamic and horizon checks:
  - Dynamic panel with lagged dependent variable instrumented: ln(GDP p.c.) continues to have significant negative effect on Gini and 5th quintile; positive effects on 1st–4th quintiles persist. Example: ln(GDP p.c.) = -8.85** (3.96) for Gini; Lagged Dependent Var. = 0.23*** (0.08). Kleibergen Paap F-Stat examples: 19.17; Hansen J p-values often > 0.3.
  - Lead and lag effects: lead effects insignificant and small; lag effects insignificant; contemporaneous ln(GDP p.c.) similar to baseline (example: ln(GDP p.c.) = -8.05** (4.03); lead = -1.45 (3.57)).
  - Longer horizons (10, 15, 20 years) yield somewhat larger absolute effects:
    - 10 years: ln(GDP p.c., 10 years) = -8.99** (4.08)
    - 15 years: -10.25** (4.37)
    - 20 years: -11.63** (5.53)
- Alternative inequality datasets:
  - Solt (2009) Net Gini (2SLS): ln(GDP p.c.) = -9.20*** (2.30)
  - Solt (2009) Gross Gini (2SLS): ln(GDP p.c.) = -14.57*** (4.32)
  - WDI (2013) Gini (smaller sample, 226 observations): ln(GDP p.c.) = -21.07* (12.75)
    - WDI quintile results: positive significant effects on 1st and 2nd quintiles; third/fourth quintiles positive but insignificant; fifth quintile negative and insignificant.
- Outliers and sample exclusions:
  - Excluding top 5th percentile changes leaves second-stage estimates largely unchanged.
  - Excluding large oil importers/exporters (Canada, France, Iran, Japan, Norway, USA, United Kingdom, South Korea, Venezuela): 2SLS with lagged dependent variable yields ln(GDP p.c.) = -9.93** (4.00) on Gini; Kleibergen Paap F-Stat: 18.45; Hansen J p-value example: 0.79.

### Heterogeneity by development level and region
- Development-level interactions (Table 10):
  - Interaction tests mostly insignificant: cannot reject that effects of national income on Gini are the same between high and low income countries.
  - Example: ln(GDP p.c.) = -8.57** (4.18); ln(GDP p.c.)*Average GDP p.c. = 0.02 (0.08).
  - Interaction with indicator for above-median average Gini significant:
    - Baseline ln(GDP p.c.) = -11.84** (4.70); interaction coefficient = 2.08** (1.02).
    - Implied marginal effects: below-median -9.76 (SE 4.17); above-median -11.84 (SE 4.70).
- Regional differences (Table 11):
  - Baseline [A] ln(GDP p.c.) = -15.25** (6.78) when region interactions included.
  - Latin America interaction [C] = 3.01** (1.15) → implied Latin America effect [A]+[C] = -12.25 (sum reported; standard error 7.12; significance at 10 percent).
  - Asia interaction [D] = 4.61** (1.86) → implied Asia effect [A]+[D] = -12.25* (7.12).
  - Sub-Saharan Africa interaction [B] = -3.75 (9.75) — not significant.
  - Quintile-specific patterns: Latin America and Asia show smaller positive effects on lower quintiles and larger positive effects on upper quintile relative to baseline.

### Mechanisms and channels — evidence and magnitudes
- Education channel (Tables 12–13):
  - Conditioning on education substantially reduces the coefficient on national income.
    - Example: conditional ln(GDP p.c.) = -6.32 (4.10) controlling for average years of education vs unconditional ln(GDP p.c.) = -8.48** (4.09).
    - Share of population with secondary education coefficient on Gini: -0.23*** (0.07) in Panel A columns 5–8.
  - Income raises education (Panel C, Table 12):
    - Average years of education: ln(GDP p.c.) coefficients 1.84*** (0.39) over 5 years up to 2.10*** (0.47) over 15 years.
    - Share with secondary education: ln(GDP p.c.) coefficients 16.50*** (4.38) up to 23.28*** (6.56) over 20 years.
    - Reported quantitative interpretations: doubling national income → approx increase in average years of education of around 2 years; doubling national income → approx increase in share with secondary education by around 20 percentage points.
  - Instrumenting both income and education (Table 13):
    - Panel A (average years instrumented): Kleibergen Paap F-Stat 11.70; Average Years of Education effect on Gini = -2.46*** (0.93); ln(GDP p.c.) becomes statistically insignificant in many specifications (example ln(GDP p.c.) = -3.95 (4.20) on Gini).
    - Panel B (share with secondary instrumented): Kleibergen Paap F-Stat 10.17; Share of Population Secondary Education effect on Gini = -0.29*** (0.09); ln(GDP p.c.) generally insignificant.
  - Conclusion: education is the most important channel mediating the income growth–inequality relationship among channels examined.
- Financial development and government size (Table 14):
  - Domestic credit/GDP coefficient: 0.04** (0.02) in Column (1) (positive association with Gini); becomes insignificant when secondary education included (Column (2): 0.02 (0.02)).
  - Government expenditures/GDP coefficient: -0.55* (0.32) in Column (3); becomes -0.37 (0.33) when secondary education included (Column (4), insignificant).
  - In specifications including share with secondary education, ln(GDP p.c.) becomes quantitatively smaller and statistically insignificant, reinforcing education's primacy.

### Key descriptive statistics (Appendix Table 1)
- Gini: Mean 39.05; Stdv 10.34; Observations 311
- 1st Quintile: Mean 6.50; Stdv 2.16; Observations 311
- 2nd Quintile: Mean 10.84; Stdv 2.66; Observations 311
- 3rd Quintile: Mean 15.13; Stdv 2.58; Observations 311
- 4th Quintile: Mean 21.46; Stdv 2.21; Observations 311
- 5th Quintile: Mean 46.10; Stdv 8.86; Observations 311
- ln(GDP p.c.): Mean 8.11; Stdv 1.29; Observations 311

### Conclusions and policy implications
- Empirical conclusion:
  - Higher national incomes lead to lower within-country inequality; growth boosts incomes of the lowest four quintiles at the expense of the top quintile.
  - Baseline 2SLS magnitude: ln(GDP p.c.) ≈ -8.48** (4.09) on Gini.
  - Paper summary: "a one percent increase in GDP per capita, on average, leads to a reduction in the Gini coefficient of over 0.08 percentage points."
- Interpretation of recent inequality increases:
  - Factors such as skill-biased technological change may drive inequality beyond GDP effects and bias simple OLS estimates.
- Policy recommendations emphasized:
  - Education policies that promote equity are potent levers to reduce income disparities and improve intergenerational earnings mobility.
  - For advanced economies: promote upper secondary or tertiary education.
  - For countries with low education attainment: promote equal access to basic education (examples cited: cash transfers to encourage attendance, public spending on education/public capital benefiting the poor).
  - Future research: more granular analysis of fiscal policy (redistributive tax/expenditure policies), health access, employment, and safety nets as mediators of how income growth affects inequality.

*Source: _wp14101 - 1. Effects of National Income on National Income Inequality (Baseline).*

### 1. Effects of National Income on National Income Inequality (Baseline) ...............................22

### 1. Effects of National Income on National Income Inequality (Baseline)

### Summary of research question and approach
- Core questions:
  - Is income inequality reduced in the context of economic growth?
  - Does the relationship depend on the country’s level of economic development?
  - Are there non-linearities (e.g., a Kuznets inverted-U)?
  - What policy channel(s) mediate the inequality-reducing effect of income growth?
- Identification strategy:
  - Use two instrumental variables (IVs) to isolate exogenous variation in national income over five-year horizons for a large panel spanning 1960–2007.
  - IV1: oil price shocks (OPS) = Δln(OilPrice)_t * θ_c, where θ_c is the time-invariant average share of net oil exports in GDP for country c.
  - IV2: trade-weighted world income (TWWI) constructed as ∑_{j≠i} w_ij GDP_jt, where w_ij is the share of trade between country i and country j in the GDP of country i (measured using trade shares between 1980 and 1989).
- Main estimation: country fixed effects with year fixed effects using 5-year non-overlapping panel data; baseline sample period 1960–2007 and the largest possible 5-year non-overlapping sample comprises a total of 80 countries.

### Principal findings (causal effects from national income to inequality)
- Average effect:
  - A one percent increase in GDP per capita reduces the Gini coefficient by around 0.08 percentage points.
- Distributional pattern:
  - Income growth boosts the relative income share of the lower quintile and the middle-class.
  - Income growth results in a decline in the income share accruing to the top quintile.
- Poverty:
  - Increases in income are associated with a significant decrease in the poverty gap and headcount ratios.
- Example (preserving exact figures reported):
  - "For example, a country like Malaysia, with a (sample average) Gini coefficient of 0.46 and (sample average) PPP GDP per capita of $5174, would expect to see its Gini coefficient decline to 0.38 as its income per capita doubles."
- Robustness:
  - Results remain intact across alternative estimation techniques, different time periods and horizons, sensitivity checks to outliers, and alternative data sources.
- Non-linearity / Kuznets curve:
  - No support found for a Kuznets curve; within-country income inequality declines as the country develops.
- Heterogeneity:
  - The inequality-reducing effect of income growth holds across countries at different stages of development and across regions.
  - Low levels of economic development and a high share of rural and agricultural populations weaken the inequality-reducing effect, but this difference is not statistically significant.
  - The inequality-reducing effects are significantly smaller in Latin American and Asian countries (growth leads to smaller income gains for the lower quintiles) but remain significantly different from zero.

### Mechanisms and channels
- Education:
  - Including education as a control substantially reduces the coefficient on national income.
  - Increases in income have a significant positive effect on educational attainment.
  - Education is identified as the most important channel mediating the income growth–inequality relationship, more so than government expenditures per se or financial sector development.
- Theoretical motivation:
  - Cites Galor and Zeira (1993): with credit market imperfections and indivisibilities in human capital investment, aggregate income increases are associated with lower inequality; education is a key channel.
- Other channels discussed in literature review:
  - Technological and structural change (skill-biased vs. not).
  - Globalization and off-shoring effects on labor demand.
  - Increased state capacity following income windfalls (tax revenues, public education, social insurance).

### Data sources and key sample statistics
- Inequality data:
  - Primary source: UN-WIDER World Income Inequality Database (WIID, 2008), filtered to drop low-quality observations.
  - Supplemented with World Bank’s POVCALNET for developing countries.
  - Initial total: 3319 survey-years for 154 countries spanning 1960–2007.
  - After cleaning (dropping duplicates, inferior quality WIID duplicates, survey-years missing extra information, and survey-years where income shares do not sum to between 99 and 101): 2785 survey-years for 144 countries.
  - Inequality data aggregated to the 5-year level by simple averages.
  - Regression focus: countries with inequality data for at least two or more consecutive 5-year intervals.
- Inequality measures:
  - Gini coefficient.
  - Income shares of the first, second, third, fourth, and fifth quintiles.
- National income:
  - Real GDP per capita from the Penn World Table (Heston et al., 2012).
- Instruments and other variables:
  - OPS data from Brueckner et al. (2012).
  - TWWI data from Acemoglu et al. (2008) using trade shares between 1980 and 1989.
  - Education: average years of education (15+) and share with completed secondary school from Barro and Lee (2010).
  - Financial development: ratio of domestic credit to the private sector over GDP from WDI (2013).
  - Government consumption share from Penn World Table (Heston et al., 2012).
- Final usable cross-country 5-year sample given availability of all required variables: 80 countries.

### Estimation framework (baseline regression)
- Baseline specification:
  - Inequality_it = α_i + β_t + η ln(GDP p.c.)_it + u_it
  - α_i : country fixed effects (control for time-invariant determinants).
  - β_t : year fixed effects (control for common shocks).
- Rationale for levels specification:
  - With country fixed effects, using the level of GDP per capita addresses how the within-country distribution of income changes as log national income changes.

*Source: _wp14101 - 1. Effects of National Income on National Income Inequality (Baseline).*

### Appendix Table 1 for descriptive statistics and Appendix Table 2 for the list of countries in

### _wp14101 - Appendix Table 1 for descriptive statistics and Appendix Table 2 for the list of countries in

### Endogeneity and 2SLS identification
- First-stage equation used to correct for endogeneity biases (2SLS):
  - (2)              ln(GDPp.c.)_{it} = a_i + b_t + c ln(TWWI)_{it} + d OPS_{it} + e_{it}
- Exclusion restriction assumption:
  - Trade-weighted world income (TWWI) and oil price shocks (OPS) affect income inequality only through their effect on GDP per capita.

### Baseline empirical results (Panels A and B of Table 1)
- Estimation methods:
  - Panel A: 2SLS estimation (controls: country and time fixed effects).
  - Panel B: LS estimation (controls: country and time fixed effects).
- Dependent variables:
  - Column (1): Gini coefficient.
  - Columns (2)–(6): income shares accruing to the 1st–5th quintiles.
- Main 2SLS finding:
  - Exogenous within-country variations in GDP per capita are negatively and significantly related to within-country variations in income inequality (Table 2).
- Quantitative interpretation (example from column (1)):
  - Estimated coefficient on log GDP per capita = -8.5.
  - Interpretation: a one percent (0.01 log point) increase in GDP per capita leads to a reduction in the Gini coefficient of over 0.08 percentage points.
  - Numerical example: Philippines — sample average Gini coefficient = 0.43; sample average PPP GDP per capita = $2769; as income per capita doubles, the Gini coefficient would be expected to decline to 0.35.
- Quintile-specific effects (2SLS):
  - Within-country increases in GDP per capita have positive effects on the income shares accruing to the 1st, 2nd, 3rd, and 4th quintiles.
  - Within-country increases in GDP per capita have negative effects on the income share accruing to the 5th quintile.

### LS estimates, Kuznets relationship, and interpretation differences
- LS results (Panel B):
  - LS estimates yield an insignificant relationship between national income and income inequality when controlling for country and year fixed effects.
- Panel C (level and squared GDP per capita to test Kuznets):
  - Inclusion of GDP per capita level and a squared term yields insignificant results.
  - The p-value from a test that the coefficients on the linear and squared GDP per capita term are equal always exceeds 0.1.
  - Conclusion: no evidence of a Kuznets relationship in the data.
- Reasons for discrepancy between 2SLS and LS estimates:
  - Measurement error: classical measurement error attenuates LS estimates toward zero but not IV estimates.
  - Reverse causality: inequality → GDP per capita could bias LS estimates.
  - Omitted variables could lead to inconsistent LS estimates; under instrument exogeneity, omitted variables would not bias 2SLS estimates.

### Instrument diagnostics
- Instruments' relevance:
  - First-stage Kleibergen-Paap F-statistic is well in excess of 10 in all regressions.
  - Implication: reject weak instrument bias according to the criterion in Staiger and Stock (1997).
- Instrument validity:
  - Hansen J test reported to examine instrument validity (joint test described; detailed results not shown in the supplied excerpt).

*Source: _wp14101 - Appendix Table 1 for descriptive statistics and Appendix Table 2 for the list of countries in*

### Appendix Table 3 shows the first stage effects of OPS and TWWI on countries' real GDP per capita. As

### _wp14101 - Appendix Table 3 shows the first stage effects of OPS and TWWI on countries' real GDP per capita. As

### Instruments, identification, and validity
- Instruments: trade-weighted world income (TWWI) and oil price shocks (OPS); oil price instrument is interaction between change in international oil price and countries' oil net-export GDP shares.
- First-stage: instruments' first-stage fit is "positive and highly significant" (consistent with prior literature).
- Hansen J test: p-value "always in excess of 0.1", suggesting inability to reject instrument validity in reported specifications (examples: Table 1 Panel A Hansen J, p-value 0.47; Tables report other Hansen J p-values across regressions).
- Exclusion restriction concern: tested correlation between countries' average oil net-export GDP shares and Gini coefficient: correlation coefficient 0.11; regression yields t-value 1.01—statistically cannot reject zero cross-country relationship.
- Conditional tests: effects of the oil price instrument on income inequality conditional on GDP per capita are insignificant (reported in Table 3), providing no evidence of a direct effect of OPS on inequality beyond GDP per capita.

### Robustness checks: specification, dynamics, horizons, outliers
- Dynamic panel with lagged dependent variable instrumented: GDP per capita continues to have a significant negative effect on Gini and the 5th quintile share; effects on 1st–4th quintiles remain significantly positive (Table 4 Panel A).
  - Example coefficients: Table 4 Panel A ln(GDP p.c.) = -8.85** (3.96) for Gini; Lagged Dependent Var. = 0.23*** (0.08).
  - Kleibergen Paap F-Stat examples: 19.17 (Gini), Hansen J p-values often > 0.3.
- Lead and lag effects (Table 5): lead effects insignificant and quantitatively small; lag effects insignificant; contemporaneous ln(GDP p.c.) remains significant and quantitatively similar to baseline.
  - Example: Panel A ln(GDP p.c.) = -8.05** (4.03); ln(GDP p.c.), lead = -1.45 (3.57).
- Longer horizons (Table 6): using ln(GDP p.c.) over 10, 15, 20 years yields somewhat larger absolute effects.
  - Panel A (10 years): ln(GDP p.c., 10 years) = -8.99** (4.08).
  - Panel B (15 years): -10.25** (4.37).
  - Panel C (20 years): -11.63** (5.53).
- First-difference estimation: similar results to levels estimation (Appendix Table 6 referenced).
- Outliers: excluding top 5th percentile positive/negative within-country changes in inequality or GDP per capita leaves second-stage estimates largely unchanged (Appendix Table 7 referenced).

### Baseline quantitative findings (selected)
- Main baseline (Table 1, Panel A, 2SLS): ln(GDP p.c.) = -8.48** (4.09) on Gini; plus quintile effects:
  - 1st Quintile: 1.43 (0.94)
  - 2nd Quintile: 1.70* (1.06)
  - 3rd Quintile: 2.01* (1.09)
  - 4th Quintile: 2.49*** (0.89)
  - 5th Quintile: -7.77** (3.52)
- Interpretation reported in conclusions: "a one percent increase in GDP per capita, on average, leads to a reduction in the Gini coefficient of over 0.08 percentage points." (paper summary statement)

### Robustness: excluding large oil importers/exporters
- Excluded countries: listed in note (Canada, France, Iran, Japan, Norway, USA, United Kingdom, South Korea, and Venezuela).
- Table 2 (2SLS with lagged dependent variable): ln(GDP p.c.) = -9.93** (4.00) on Gini; quintile effects remain positive for 1st–4th and negative for 5th.
  - Kleibergen Paap F-Stat: 18.45; Hansen J, p-value examples: 0.79 (Gini).

### Alternative inequality datasets and poverty
- Solt (2009) dataset:
  - Net Gini (2SLS): ln(GDP p.c.) = -9.20*** (2.30).
  - Gross Gini (2SLS): ln(GDP p.c.) = -14.57*** (4.32).
- WDI (2013) Gini (smaller sample, 226 observations): ln(GDP p.c.) = -21.07* (12.75) (Table 8 Column (3)).
  - WDI quintile results: positive significant effects on 1st and 2nd quintiles; third/fourth quintiles positive but insignificant; fifth quintile negative and insignificant (Table 8 columns (4)-(8)).
- Poverty (Table 9):
  - Panel A (country FE only, 2SLS): ln(GDP p.c.) effects:
    - PovGap $1.25: -3.93** (1.80)
    - PovGap $2.00: -6.57*** (2.43)
    - PovCount $1.25: -9.78** (3.90)
    - PovCount $2.00: -11.16*** (3.34)
  - Panel B (country & year FE): coefficients larger in magnitude but with much larger standard errors (e.g., ln(GDP p.c.) = -8.96 (11.24) for PovGap $1.25).
  - Kleibergen Paap F-Stat frequently low for poverty regressions: example 6.57 (Panel A), 4.19 (Panel B).

### Heterogeneity by development level and region
- Role of economic development (Table 10):
  - Interaction tests mostly insignificant: cannot reject that effects of national income on Gini are the same between high and low income countries.
  - Example: Column (1) ln(GDP p.c.) = -8.57** (4.18); ln(GDP p.c.)*Average GDP p.c. = 0.02 (0.08).
  - Quantitative illustrations given: interaction effect changing marginal effect from -8.51 to -8.36 using 25th/75th percentiles; similar calculation for rural population share.
  - Exception: interaction with indicator for above-median average Gini (column (6)) is significant: ln(GDP p.c.) = -11.84** (4.70) for baseline; interaction coefficient 2.08** (1.02). Implied marginal effects: below-median -9.76 (SE 4.17); above-median -11.84 (SE 4.70).
- Regional differences (Table 11):
  - Baseline ln(GDP p.c.) [A] = -15.25** (6.78) when region interactions included.
  - Latin America interaction [C] = 3.01** (1.15) — implies Latin America overall effect [A]+[C] = -12.25 (sum reported as -15.25 + 3.01 = -12.25 with standard error 7.12 and significance at 10 percent level).
  - Asia interaction [D] = 4.61** (1.86) — implied Asia effect [A]+[D] = -12.25* (7.12).
  - Sub-Saharan Africa interaction [B] = -3.75 (9.75) — not significant.
  - Quintile-specific patterns: Latin America and Asia show smaller positive effects on lower quintiles and larger positive effects on upper quintile relative to baseline.

### Channels: education, financial development, government size
- Education channel (Tables 12–13):
  - Conditional regressions (Table 12 Panel A): controlling for education reduces the statistical significance and magnitude of instrumented national income.
    - Example: over 5 years, conditional ln(GDP p.c.) = -6.32 (4.10) controlling for average years of education; unconditional ln(GDP p.c.) = -8.48** (4.09) (Panel B).
    - Share of population with secondary education coefficient: -0.23*** (0.07) (Panel A columns 5–8 show consistent significance).
  - Panel C (Table 12): income raises education:
    - Average years of education: ln(GDP p.c.) coefficients 1.84*** (0.39) over 5 years up to 2.10*** (0.47) over 15 years.
    - Share with secondary education: ln(GDP p.c.) coefficients 16.50*** (4.38) up to 23.28*** (6.56) over 20 years.
    - Quantitative interpretation reported: doubling national income leads to approx increase in average years of education of around 2 years; doubling leads to approx increase in share with secondary education by around 20 percentage points.
  - Instrumenting both income and education (Table 13):
    - Panel A (average years instrumented): Kleibergen Paap F-Stat 11.70; Average Years of Education effect on Gini = -2.46*** (0.93); ln(GDP p.c.) becomes statistically insignificant in many columns (e.g., ln(GDP p.c.) = -3.95 (4.20) on Gini).
    - Panel B (share with secondary instrumented): Kleibergen Paap F-Stat 10.17; Share of Population Secondary Education effect on Gini = -0.29*** (0.09); ln(GDP p.c.) again generally insignificant.
  - Overall conclusion: education is a key channel; conditioning on education reduces or eliminates the statistical effect of income on inequality.
- Financial development and government size (Table 14):
  - Domestic credit/GDP coefficient: 0.04** (0.02) in Column (1) (positive association with Gini); becomes insignificant when secondary education included (Column (2): 0.02 (0.02)).
  - Government expenditures/GDP coefficient: -0.55* (0.32) in Column (3); becomes -0.37 (0.33) when secondary education included (Column (4), insignificant).
  - In regressions including share of population with secondary education, ln(GDP p.c.) becomes quantitatively smaller and statistically insignificant, reinforcing education's primacy as the mediating channel.

### Conclusions and policy implications
- Main empirical conclusion: higher national incomes lead to lower inequality; growth boosts incomes of the lowest four quintiles at the expense of the top quintile.
  - Baseline 2SLS magnitude: ln(GDP p.c.) ≈ -8.48** (4.09) on Gini (Table 1 Panel A).
  - Paper summary: "a one percent increase in GDP per capita, on average, leads to a reduction in the Gini coefficient of over 0.08 percentage points."
- Interpretation of recent increases in inequality: may be driven by factors like skill-biased technological change that disproportionately benefit higher-skilled workers and affect inequality beyond GDP effects; such factors would bias simple OLS but not the instrumented regressions.
- Policy implications emphasized:
  - Education policies that promote equity can be potent levers to reduce income disparities and improve intergenerational earnings mobility.
  - For advanced economies: promote upper secondary or tertiary education.
  - For countries with low education attainment: promote equal access to basic education (e.g., cash transfers to encourage attendance, public spending on education/public capital benefiting the poor).
  - Future research suggested: more granular analysis of fiscal policy (redistributive tax/expenditure policies), health access, employment, and safety nets as mediators of how income growth affects inequality.

*Source: IMF working paper content unit _wp14101 (Appendix excerpts and tables as provided).*

### Appendix Table 1. Descriptive Statistics

### Appendix Table 1. Descriptive Statistics

### Key descriptive statistics (means, standard deviations, observations)
- Gini: Mean 39.05; Stdv 10.34; Observations 311
- 1st Quintile: Mean 6.50; Stdv 2.16; Observations 311
- 2nd Quintile: Mean 10.84; Stdv 2.66; Observations 311
- 3rd Quintile: Mean 15.13; Stdv 2.58; Observations 311
- 4th Quintile: Mean 21.46; Stdv 2.21; Observations 311
- 5th Quintile: Mean 46.10; Stdv 8.86; Observations 311
- ln(GDP p.c.): Mean 8.11; Stdv 1.29; Observations 311

*Source: Appendix Table 1. Descriptive Statistics*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14101.pdf_